A dynamic steady-state model of financial reporting subject to ongoing earnings management strategies is presented. I show that, even in the absence of researcher measurement error, discretionary accruals and nondiscretionary accruals are always negatively correlated. This is inconsistent with the currently predominant least-squares approach to estimating these components of earnings. This negative correlation becomes weaker (less negative) as earnings persistence increases, and (counter-intuitively) as the ability to engage in earnings management increases. These results provide some building-blocks for developing improved technology for empirical investigation of earnings management using discretionary accruals.